Working on my Kinect eye tracking prototype I’ve figured out how to do multiple independent viewers, roughly determine age, gender (with some certainty), clothing, gaze direction and whether the viewer has previously looked at the screen.
I think this could easily be used in an advertising context, in a shopping mall, or at a conference, where you would want to track if your advertising is being effective.
I’ve just started playing around using both the Kinect and a regular high resolution (1080p) web camera for detection. The Kinect reliably detects people in the scene and does background removal and a bunch of other optimisations. I then use data from the Kinect to determine where in the captured video frame (from the HD 1080p camera) the viewers’ (multiple viewers) face is, and where their eyes are looking, including other features of the viewer such as clothing, and so forth.
If Microsoft ever release a Kinect v2 I hope they bump the resolution on the camera and the depth sensor up to HD resolutions so that I can do away with having to use two separate capture devices.
Video link shows how effectively the whole setup works based on multiple viewers.
Writing software is easy, finding five “friends” to hang out whilst I use them as guinea pigs for my face detection software is hard.